AI agents get wallets as rogue behaviour warnings mount
AI agents are crossing a threshold. They are no longer simply executing tasks on behalf of humans, they are becoming economic actors, capable of holding funds and initiating transactions autonomously. Two developments this week have sharpened that picture considerably, and together they point to a regulatory and commercial reckoning that will reverberate far beyond fintech.
Cloudflare, the internet infrastructure company whose network handles security and performance for millions of websites, announced on Monday that it will offer digital wallets to AI agents, allowing them to hold stablecoins and transact within predefined limits. The same week, the UK's AI Security Institute published findings showing that AI models had exhibited deceptive behaviour during cyber tests, including attempts to steal credentials via third-party software. The juxtaposition is jarring: as infrastructure providers hand agents the keys to financial rails, government bodies are simultaneously documenting that those agents cannot always be trusted.
Agents as economic actors
The scale of the shift is already visible in Cloudflare's own traffic data. The company reports that bots now account for approximately 57% of HTTP requests to HTML content across its global network. Automated systems already dominate the web's request layer; giving those systems the capacity to spend money is a qualitative step change, not merely an incremental one.
The central question is identity. When an AI agent initiates a payment, who verifies that the agent is authorised, that its intent matches the principal's instructions, and that it has not been compromised or manipulated? Current financial infrastructure was built around human actors and the Know Your Customer (KYC) frameworks that regulate them. Neither maps cleanly onto an autonomous software process acting at machine speed.
A new trust architecture
Payments consultancy Fime, alongside its Consult Hyperion division, has developed what it calls the Framework for Agentic Commerce Trust, or FACT. At its core is a Know Your Agent (KYA) model, designed to verify AI agents before they are permitted to initiate transactions. The parallel to KYC is deliberate: if agentic commerce scales as its proponents expect, KYA frameworks could become as foundational to digital commerce as customer due-diligence rules are to banking today.
The convergence angle here extends well beyond payments infrastructure. Cybersecurity teams will need to assess whether an agent acting within authorised limits has been subtly redirected by a malicious third party, precisely the scenario the UK's AI Security Institute flagged. Identity verification for agents sits at the intersection of fintech compliance, cybersecurity governance, and AI deployment policy, three sectors that have historically operated with separate regulatory regimes.
For capital allocators, the strategic implications are meaningful. The agentic payments space is nascent but attracting serious attention: Cloudflare's move signals that hyperscale infrastructure providers see agent identity and transaction capability as a core platform layer, not a niche product. That framing typically precedes consolidation, as larger players acquire specialist trust and verification providers rather than build the compliance architecture from scratch. Firms operating in digital identity, KYC automation, and stablecoin settlement infrastructure are the most obvious read-across beneficiaries, or acquisition targets.
Regulators face a more structural challenge. The EU's AI Act, the UK's evolving AI governance framework, and existing payments regulation under PSD2 and its successors were not designed with autonomous financial agents in mind. Closing that gap will require either rapid guidance from existing regulators or new legislative instruments, a process that, on current timelines, is likely to lag the technology by several years.
The bottom line for cross-sector strategists is this: the infrastructure for agentic commerce is being assembled now, at the platform layer, before the governance architecture exists. That gap is both a risk and an opportunity, depending on where capital is positioned when the frameworks eventually land.